Discussion of: What Undermines Aid s Impact on Growth? by Raghuram Rajan and Arvind Subramanian Aart Kraay The World Bank Presented at the Trade and Growth Conference, Research Department Hosted by the International Monetary Fund Washington, DC January 9, 2006 The views expressed in this paper are those of the author(s) only, and the presence of them, or of links to them, on the IMF website does not imply that the IMF, its Executive Board, or its management endorses or shares the views expressed in the paper.
Plan for Discussion Simple overview of RS story Mapping of theory to data Focus on structural equations in RS story Econometrics Big Picture: what have we learned and should we worry about it?
1. Rajan-Subramanian Story: Aid Overvaluation Dutch Disease Aid leads to overvaluation: Over(j) = γ Aid(j) + v(j), γ>0 Overvaluation leads to slower (faster) growth in labour- (capital-) intensive sectors: RelGrow(j) = β Over(j) + e(j), β<0 Note: RelGrow is defined as growth of all labour-intensive industries relative to all capital-intensive industries (one observation per country)
Stylized Version of RS, Cont d Structural Model: Over(j) = γ Aid(j) + v(j), γ>0 RelGrow(j) = β Over(j) + e(j), β<0 Reduced-Form Model: RelGrow(j) = βγ Aid(j) + {βv(j) + e(j)} RS estimate Reduced-Form Model using IV concern is that aid is correlated with other determinants of overvaluation, CORR(Aid,v) 0 use clever instrument Z from other paper exclusion restriction: E[Z(βv+e)]=0 (non-trivial!)
2. Mapping Theory to Data Inflows of foreign aid bid up costs of factors of production Adverse effect on tradeable sector which faces fixed world prices and can t pass on higher costs Tradeables production should grow more slowly than non-tradeables production in countries with lots of aid
Theory to Data, Cont d RS argue we can t observe which sectors are tradeable Or can we? Most services are non-traded, most manufacturing is traded Is growth of manufacturing relative to services lower in countries that get lots of aid? construct analog of RS Figure 3 using growth of manufacturing relative to services correlation goes wrong way more aid leads to faster growth in manufacturing relative to services
My Version of Chart 3: Manufacturing = Tradeables, Services = Non-Tradeables 0.06 g(manufacturing) - g(services) 0.04 0.02 0-0.02-0.04-0.06-0.08 Lesotho Swaziland Indonesia Grenada Seychelles Nepal Central BeninBhutan African Cameroon Malaysia Republic Tonga Comoros Bangladesh Cote Sri d'ivoire Lanka Zambia Togo Mauritius Fiji Senegal Dominican Costa Rica Gambia, The Congo, Dominica Papua Rep. Honduras New Pakistan Jordan 0 Malawi Republic Tunisia Morocco Guinea Kenya Mali Nicaragua Paraguay Peru St. Lucia Namibia El Salvador Bolivia Nigeria Antigua Ecuador Panama Guatemala Philippines Zimbabwe and Jamaica Belize Barbuda Botswana St. Kitts and St. Vincent Burkina and Faso Nevis the Grenadines Rwanda Suriname Haiti Ghana y = 0.00x - 0.00 R 2 = 0.04 0 5 10 15 20 25 30 35 40 Aid/GDP in 1990
From Theory to Data, Cont d Instead of identifying tradeable sectors, RS identify sectors that are labour-intensive Key assumption: labour is the factor whose price is bid up by aid Implication: labour-intensive tradeable sectors will grow relatively slowly in countries with lots of aid Do aid inflows simply bid up the price of labour? skilled versus unskilled labour: e.g. aid agencies poach highly-skilled administrators, translators, (economists?) etc. from government or private sector labour versus capital: e.g. aid agencies monopolize all the 4-wheel drive vehicles
Another Version of Chart 3: Manufacturing = Skill- Intensive, Agriculture = Unskilled-Intensive 0.12 g(manufacturing) - g(agriculture) 0.1 0.08 0.06 0.04 0.02 0-0.02-0.04-0.06 Grenada Malaysia Botswana Swaziland Mauritius Seychelles Indonesia LesothoBhutan Nepal St. Kitts Sri and Lanka Dominican Tonga Nevis Antigua Republic St. Morocco Lucia and Bangladesh Dominica Barbuda Fiji El Salvador Pakistan Senegal Benin Central African Cameroon Costa Rica Cote Congo, d'ivoire Rep. Kenya Jordan Tunisia Honduras Ecuador Philippines Namibia Zambia Comoros Republic Mali Nigeria Guatemala Jamaica St. Vincent Bolivia Papua and Togo New Paraguay Panama Sudan Belize Guinea Zimbabwe the Grenadines Malawi Peru Haiti Ghana Burkina Faso y = -0.00x + 0.02 R 2 = 0.01 Gambia, The Nicaragua Rwanda Suriname 0 5 10 15 20 25 30 35 40 Aid/GDP in 1990
3. Focus on Structural Equations Structural Model: Reduced-Form Model: Over(j) = γ Aid(j) + v(j), γ>0 RelGrow(j) = β Over(j) + e(j), β<0 RelGrow(j) = βγ Aid(j) + {βv(j) + e(j)} RS spend most of paper trying to estimate the reducedform parameter βγ using IV (i.e. all of Tables 2-7) What about two key structural equations?
Structural Equation 1: Aid and Overvaluation Over(j) = γ Aid(j) + v(j): RS offer us only one univariate IV regression described briefly in text p.25, and Figure 3 Big(gish) previous literature on aid an overvaluations has looked at this with cross-country data (surveyed by Adam (2005), Bulir and Lane (2002)) existing evidence is pretty mixed at most small effect, doubling aid leads to 18% appreciation over 5 years (Prati et al (2003))
Aid and Overvaluation, Cont d Why Do RS find a big effect? level versus rate of change of RXR? omitted variables correlated with aid (commodity dependence, institutions, terms of trade shocks)? mechanical correlation running through Balassa Samuelson correction? RS Two-Step: p = ηy + e, ehat = γaid + v Effect of Aid on RXR: p = ηy + φaid + u Two methods are NOT identical since CORR(Aid,y)<*0 Implication is that φ<< γ
Aid and Overvaluation All Countrys RS Sample for 1990s Dep Variable Overvaluation Price Level Overvaluation Price Level Aid -0.08-0.10 1.71 1.29 (0.19) (0.22) (0.79)** (0.81) Per Capita GDP 0.002 0.001 (0.0005)*** (0.001) # Countries 68 68 15 15 Not clear that even the RS measure of overvaluation is correlated with aid in larger sample of aid recipients Relevant question is: does aid raise RXR? evidence suggests not (cols 2 and 4)
Structural Equation 2: Overvaluation and Relative Growth of Labour-Intensive Sectors RelGrow(j) = β Over(j) + e(j) This is main novelty of paper, but we see it only Table 8 Result a bit puzzling: why should overvaluation disproportionately affect labour-intensive sectors? Results much stronger if we instead have relative growth of export-intensive sectors Direct evidence on how much overvaluation hurts overall manufacturing or export growth in a big sample of countries would be more convincing
Effects of Overvaluation, Cont d More important issues is effect of overvaluation on overall manufacturing growth RS do this (Table 9), find strongly negative effect!!! Is this really robust? not really effect of aid on average manufacturing growth, but on unweighted average of sectoral growth artifact of small sample and/or unclustered standard errors? Concerns about composition of RS sample Few really poor countries (only 3 of 28 HIPCs!) Not very aid-dependent (average aid/gdp in 1990s sample is 3.6%, for all countries it is 7.6%)
Aid and Manufacturing Growth in 1990s 15 Manufacturing Growth in 1990s 10 5 0-5 -10 Uganda Lao PDR Vietnam Malaysia Syrian Arab Republic Nepal Seychelles Sri Lanka Indonesia Costa Grenada Rica Bangladesh Maldives Egypt, Lesotho Benin Arab Bhutan Rep. Dominican Tunisia Mauritius Republic El Salvador Papua New Guinea Jordan Botswana St. Kitts Belize and Nevis Guinea Honduras Peru Pakistan Sudan Cote d'ivoire Bolivia Senegal Cape Nicaragua Verde Panama Guatemala Fiji Philippines Morocco Namibia Swaziland Yemen, Rep. Cameroon Dominica Madagascar Kenya Togo Tonga Tanzania Niger Comoros Nigeria Vanuatu Burkina FasoZambia Paraguay Central African Gambia, The Antigua Ecuador and Barbuda St. Vincent Ethiopia and the Mauritania Malawi Angola St. Lucia Republic Zimbabw e Grenadines Mali Jamaica Congo, Rep. Ghana Suriname Rwanda Haiti y = -0.02x + 3.35 R 2 = 0.00 0 5 10 15 20 25 30 35 40 Aid/GDP in 1990
4. Econometric Issues: Clustering Dependent variable is growth of sector i in country j Country (industry) dummies soak up effects of country (industry) shocks only if all industries in a country (countries in an industry) respond the same way to the shock Correcting standard errors for correlations across countries and across industries is very important RS do so for countries and industries separately, both corrections substantially increase standard errors Significance will probably fall a lot if you correct for both at the same time
Big Picture 1: Overview of RS Claims and Evidence RS: real overvaluation lowers growth in labour-intensive industries nice application of RZ methodology, intuitive result RS: aid leads to real overvaluations not yet convincing, probably artifact of small sample and/or definition of overvaluation existing literature at most weakly supportive RS: aid slows growth in labour-intensive industries and in overall manufacturing not yet convincing because of small sample problems peculiar because of missing link from aid to overvaluation question posed in title remains to be answered
Big Picture 2: How Much Should We Care About Aid-Induced Real Appreciations Anyway? Aid accounts at most for a small share of variation in RXR policy implication: give lots of aid, and address other fundamental sources of overvaluation Aid effect on RXR likely to be temporary anyhow supply responses in non-traded sector improvements in human capital of poached workers Are manufactured exports really the engine of growth? depends on how much we believe stories about externalities